A method and system for identifying precursors of coal and rock instability based on fracture topology parameters

Abstract

This invention discloses a method and system for identifying precursors of coal and rock instability based on fracture topology parameters. The method includes: acquiring acoustic data through microseismic monitoring, and extracting core parameters of coal and rock fracture by combining source location and source mechanism inversion; considering the geometric size and directional characteristics of fractures, constructing a fracture network topology model based on graph theory, and obtaining fracture spatiotemporal topological evolution parameters to construct a multivariate time-series topological feature dataset; introducing the multivariate time-series graph model MTAD-GAT, using graph attention mechanism to dynamically capture the structural dependencies between various topological parameters, mining their collaborative anomaly features, and combining with a time-series modeling module to capture their evolution trend; fusing prediction error and reconstruction error to form an anomaly score as an instability precursor index, extracting the covariance features and evolution trends of multiple key topological indicators in the precursor stage, assisting in the judgment of critical instability states, thereby achieving unsupervised automatic identification and intelligent early warning of topological precursors.

Description

Technical Field

[0001] This invention relates to the field of early warning of coal and rock instability, and in particular to a method and system for identifying precursors of coal and rock instability based on fracture topology parameters. Background Technology

[0002] As coal mining depths continue to increase, the geological structures of mines are becoming increasingly complex, with coal and rock masses subjected to long-term high-stress, multi-source disturbances, and nonlinear loading environments. This complex stress evolution process easily triggers unstable expansion and sudden rupture of internal fissures in the coal and rock mass, thereby inducing typical dynamic disasters such as rockbursts, coal and gas outbursts, roadway instability, and roof collapse. Before instability and rupture, coal and rock masses often undergo a series of microstructural evolution and energy release processes, exhibiting certain precursory characteristics. Therefore, identifying these hidden precursory signs before disasters occur and promptly determining whether the coal and rock mass has entered the instability evolution stage has become a key issue in the current field of coal mine disaster prevention and control.

[0003] Microseismic monitoring, as a highly sensitive and non-destructive detection method, can capture elastic wave signals released by micro-fracture activity in coal and rock masses under load in real time. By comprehensively analyzing the spatiotemporal distribution, frequency characteristics, and energy evolution of microseismic events, precursory features of coal and rock fractures can be effectively identified, thereby achieving early perception and accurate warning of dynamic disasters. Patent CN114895352A proposes a method and device for predicting rock mass instability based on microseismic monitoring. It constructs a logarithmic potential energy ratio based on source parameters, volumetric potential, and energy release, and establishes a fuzzy warning time model in conjunction with cumulative apparent volume to achieve time-based warning of rock mass instability. Patent CN117932371A discloses a spatiotemporal prediction method for rockburst based on the "time-space-intensity" characteristics of microseismic data. It uses an LSTM network to predict the occurrence time of rockburst based on precursory pattern sequences from microseismic data, and utilizes multi-scale spatial cloud map fusion to achieve spatial identification of hazardous areas using microseismic spatiotemporal characteristic indicators. Patent CN117150911A proposes a method and system for predicting coal and rock instability and fracturing based on graph neural networks. This method constructs a network graph by dividing microseismic signals into temporal and spatial parts, predicts future topological evolution through graph embedding and neural networks, and identifies important nodes as precursors to fracturing, achieving accurate early warning. These methods can, to some extent, reflect the energy release and disturbance response of rock masses during fracturing, and, combined with models such as neural networks, can identify precursor patterns and predict them temporally and spatially. However, they still have significant limitations: they mainly focus on the spatiotemporal distribution, frequency characteristics, and energy parameters of the microseismic signals themselves, starting only from the "signal level," failing to deeply explore the evolutionary laws of fracture structures reflected behind microseismic events, and making it difficult to comprehensively reflect the complex fracture behavior during the fracturing evolution stage. Summary of the Invention

[0004] This solution addresses the problems and needs raised above by proposing a method and system for identifying precursors of coal and rock instability based on fracture topology parameters. The above technical objectives are achieved by adopting the following technical features, and several other technical benefits are also brought about.

[0005] One objective of this invention is to propose a method for identifying precursors of coal and rock instability based on fracture topology parameters, comprising the following steps:

[0006] S10: Raw acoustic data is acquired through a microseismic monitoring system. Combined with source location calculation and source mechanism inversion, core physical parameters directly related to the micro-fracture source are obtained during the coal and rock fracturing process.

[0007] S20: Taking into account the geometric size and directional characteristics of fractures, four types of topological relationships are defined, the fracture penetration index (BCI) is quantitatively calculated, the fracture network is abstracted into a graph model based on graph theory, a topological model that characterizes the spatiotemporal evolution of fractures is constructed, multidimensional time-series topological parameters are extracted, and a multivariate topological time series dataset reflecting the evolution of coal and rock fractures is obtained.

[0008] S30: Divide the constructed multivariate topological time series dataset into training and testing sets, and perform data normalization and outlier handling on the time series of each dimension of topological structure parameters.

[0009] S40: Based on the preprocessed training set data, a multivariate time series graph model MTAD-GAT is constructed to form a dual graph attention mechanism that is feature-oriented and time-oriented, respectively modeling the structural dependencies and temporal evolution trends between topological parameters; multi-head attention is introduced into the graph attention layer, and different attention scores are calculated through multiple independent attention heads; the local and global features extracted from the graph attention layer and the one-dimensional convolutional layer are fused and input into the gated recurrent unit (GRU) to deeply capture the dynamic evolution patterns and long-term dependencies of topological features; during training, the model is iteratively optimized by minimizing the loss function and combining backpropagation and gradient update mechanisms;

[0010] S50: The trained model is applied to the test set data. The output of the gated recurrent unit (GRU) is fed into the prediction module for predicting the observations at the next time step and the reconstruction module for characterizing the overall temporal distribution features, respectively. The errors of the two modules are fused to construct an anomaly score as an instability precursor index. The temporal changes of the anomaly score are used to determine the rupture precursor interval, and then the high-contribution topological parameters and their change characteristics that change significantly within the precursor interval are identified.

[0011] Furthermore, the coal and rock instability precursor identification method and system based on fracture topology parameters according to the present invention may also have the following technical features:

[0012] In one example of the present invention, in step S10, raw acoustic data is acquired, and combined with source location calculation and source mechanism inversion, core parameters of the coal and rock fracture source directly related to the fracture are obtained. Specifically, this includes the following steps:

[0013] S11: The raw acoustic data collected by the microseismic monitoring system is post-processed, and the arrival time and amplitude information of the first wave are extracted using the red-delay information criterion method.

[0014] S12: Using the first wave arrival time data from sensors at different locations, the spatial location calculation of the acoustic emission source is performed using the simplex positioning robust algorithm;

[0015] S13: Based on the constraints of the tension-shear rupture source model, and combined with the first wave amplitude data from sensors at different locations, solve for the six components of the source rupture moment tensor.

[0016] S14: The core physical parameters of the coal and rock fracture source are quantitatively inverted by calculating the moment tensor components. The core physical parameters include the location coordinates, spatial orientation, and fracture volume.

[0017] In one example of the present invention, in step S20, four types of topological relationships are defined by comprehensively considering the geometric size and directional characteristics of the fracture, the fracture continuity index (BCI) is quantitatively calculated, the fracture network is abstracted into a graph model based on graph theory, a topological model characterizing the spatiotemporal evolution of the fracture is constructed, multidimensional time-series topological parameters are extracted, and a multivariate topological time series dataset reflecting the evolution of coal and rock fractures is obtained, including the following steps:

[0018] S21: Based on geometric relationships, using the fracture location coordinates, volume radius, and spatial orientation, construct the normal vector and parametric equations of the intersection line between any two fracture planes, and then solve them simultaneously with the equations of the sphere containing the fracture. By analyzing the spatial positional relationship of the intersection points, determine four types of topological relationships between fractures: intersecting, connected, embedded, or disjoint. Simultaneously, calculate the diameter of the intersection circle based on the distance between the fracture spherical radius and the center of the sphere to characterize the geometric connection features of the fractures.

[0019] S22: For any fractures i and j in a fracture network, after determining the existence of a connection relationship based on the spatial topological relationship between the fractures, the formula for calculating the fracture penetration index is as follows:

[0020]

[0021] In the formula, Z(V) i ) and Z(V j ) are the normalized volume values ​​of the i-th and j-th fractures, respectively, and d ij Let be the Euclidean distance between the i-th fracture and the j-th fracture;

[0022] Assuming there are n continuous fracture nodes around fracture i, the total fracture continuity index (BCI) is... i The calculation formula is as follows:

[0023]

[0024] S23: After determining the connectivity between fractures, construct a fracture network topology model based on the geometric relationship between the fracture center and the intersection center;

[0025] S24: Extract multidimensional topological attribute parameters of the fracture network, and combine them with the time series information of microseismic events to construct a multivariate topological time series feature dataset characterizing the time-varying evolution of the fracture network structure; among which, the multidimensional topological attribute parameters include: connection type, network density, clustering coefficient, cumulative openness and connectivity index.

[0026] In one example of the present invention, the process of constructing the fracture network topology model in step S23 includes the following steps:

[0027] The fracture network is abstracted into a graph structure using graph theory, where fractures are nodes and are classified into I nodes, T nodes, Y nodes, X nodes and D nodes according to the number of connecting edges. The degree of a node is defined as the total number of connecting edges.

[0028] The geometric information of the fracture is assigned to the edges connecting the nodes. The edges are divided into unidirectional edges and bidirectional edges according to the connectivity index (BCI). Among them, unidirectional edges represent lower connectivity and weak interaction, while bidirectional edges represent higher connectivity and strong interaction, which can form stable fracture channels. The fracture network topology model includes:

[0029] In one example of the present invention, step S30, which involves partitioning the multivariate topological time series dataset and performing data preprocessing, includes the following steps:

[0030] S31: Divide the constructed multivariate topological time series dataset into a training set and a test set. The training set selects data from the stable evolution stage from the early to the middle stage of coal and rock fracturing, while the test set selects data from the later stage before instability and fracturing.

[0031] S32: Normalize the time series of topological structure parameters for each dimension in the training and test sets, for topological parameters x1, x2, ..., x... n-1 x n The minimum-maximum normalization method is used, and its expression is:

[0032]

[0033] In the formula, This represents the minimum value of the topology parameter. The maximum value of the topology parameter, y i ∈[0,1];

[0034] S33: The spectral residual method is used to identify and remove outliers in the time series of topological parameters of each dimension in the training set.

[0035] In one example of the present invention, in step S40, the MTAD-GAT model includes a 1-D convolutional layer, which is used to extract high-level features for each timestamp from the input multivariate time series data, specifically including:

[0036] Let the input tensor of the multivariate time series be X∈R. T×N×D Where T is the maximum timestamp length, N is the number of variables, and D is the input feature dimension of each variable. The feature extraction formula through a 1-D convolutional layer is as follows:

[0037] X conv =Conv1D(X)∈R T×N×D′

[0038] In the formula, X conv is the output tensor of the 1-D convolutional layer, and D′ is the dimension of the local features extracted after passing through the 1-D convolutional layer.

[0039] In one example of the present invention, in step S40, the dual-graph attention mechanism includes the following steps:

[0040] S41: Linear Transformation: For each attention head m = 1, 2, ..., M, transform the output of each node after passing through a 1-D convolutional layer.

[0041] Feature X conv ∈R T×N×D′ Perform a linear mapping, the expression of which is:

[0042]

[0043] Where: h i ∈R D′ W represents the output feature of the i-th node after passing through the convolutional layer. (m) ∈R F×D′ Let F be a learnable linear mapping matrix, and let F be the output dimension of each attention head. Features after mapping;

[0044] S42: Calculating the attention coefficient: For two nodes i and j, the attention mechanism is used to calculate their similarity. The expression for the attention coefficient is:

[0045]

[0046] In the formula: Let α be the non-normalized attention score between nodes i and j in the m-th attention head, and LeakyReLU denote the non-linear activation function. (m) ∈R 2F A learnable multi-head attention vector. This represents a vector concatenation operation;

[0047] S43: Normalized Attention Weights: The attention scores are normalized using Softmax, and the expression is as follows:

[0048]

[0049] In the formula, Let represent the normalized attention score between nodes i and j in the m-th attention head, Softmax represent the activation function, and N(i) be the set of neighboring nodes of node i.

[0050] S44: Aggregate Neighbor Features: By weighting the features of neighbor nodes through attention, the expression for updating a node is:

[0051]

[0052] In the formula, σ represents the aggregated feature output of node i in the m-th attention head, where σ is a non-linear activation function.

[0053] S45: Multi-head output aggregation: Concatenates the output features of each head along its dimensions. The expression is as follows:

[0054]

[0055] In the formula, z i This represents the final representation of node i after all heads have been merged;

[0056] The final output of the feature-oriented graph attention layer is: The final output of the time-oriented graph attention layer is The outputs of the feature-oriented graph attention layer and the time-oriented graph attention layer are arranged into a unified tensor, which is then concatenated with the output of the 1-D convolutional layer to obtain the overall representation after output concatenation. This representation is then used as input to the GRU layer for dynamic temporal modeling.

[0057]

[0058] In the formula, M·F is the dimension of the final output of each GAT layer, and D″ is the total feature dimension after concatenation.

[0059] In one example of the present invention, step S50 includes the following steps:

[0060] S51: Joint Optimization Strategy to Construct Multi-Objective Loss Function: Combining the prediction and reconstruction modules, a joint loss function is introduced during training to simultaneously optimize the two sub-networks, as shown in the following formula:

[0061] Loss = Loss for +Loss rec

[0062] In the formula, Loss for Loss is the loss function for the prediction model. rec The loss function for reconstructing the model;

[0063] S52: Constructing an anomaly scoring mechanism: The model generates prediction error and reconstruction probability separately, and constructs a comprehensive anomaly scoring function by fusing the two. A peak-to-threshold algorithm is used to select the anomaly threshold on the validation set. If the anomaly score at a certain time step is greater than the anomaly threshold, that time step is marked as "abnormal"; otherwise, it is marked as "normal." The "abnormal" score is used as an instability precursor index to identify the rupture precursor interval.

[0064] The formula for calculating the comprehensive anomaly scoring function is as follows:

[0065]

[0066] In the formula, For predicted values Compared with the actual value x i The squared error between (1-p) i ) represents the probability of encountering an outlier in feature i based on the reconstructed model, and γ is a hyperparameter selected for the validation set to balance the two modules.

[0067] S53: Statistically analyze the topological evolution trend within the precursor interval: statistically analyze the characteristic quantities of each topological parameter within the precursor interval and compare them with the corresponding parameters in the normal interval; through parameter difference analysis and t-test methods, determine whether there are significant changes in the topological parameters during the precursor stage, so as to identify the high-contribution topological parameters and their change characteristics that play a key role in the rupture evolution process.

[0068] In one example of the present invention, the prediction module is configured to connect three fully connected layers with a hidden dimension of d2 after the temporal feature representation output by the gated recurrent unit (GRU) for nonlinear prediction of the data at the next time step; wherein the prediction loss function is defined using the root mean square error (RMSE) as follows:

[0069]

[0070] In the formula: x n Given the input sequence x = (x0, x1, ..., x... n-1The true value of x at the next time step. n,i Let i be the true value of the i-th feature. The predicted value is K, where K is the total number of features.

[0071] The reconstruction module is configured to learn the probability distribution of the entire input sequence in the latent space using a variational autoencoder, reconstruct the input, and calculate the reconstruction error; wherein the reconstruction loss function is expressed as:

[0072]

[0073] In the formula, the first term is the reconstruction error, which measures how close the reconstructed value is to the original input, q. φ (z|x) is the encoder. The first term is the decoder; the second term is the KL divergence, the regularized latent space distribution, and the latent variables. Represents the implicit structure of data.

[0074] Another objective of this invention is to propose a coal and rock instability precursor identification system based on fracture topology parameters, comprising:

[0075] The source parameter acquisition module is configured to acquire raw acoustic data through a microseismic monitoring system, and combine source location calculation and source mechanism inversion to obtain core physical parameters directly related to the micro-fracture source during the coal and rock fracturing process.

[0076] The topology parameter acquisition module is configured to comprehensively consider the geometric size and directional characteristics of fractures to define four types of topological relationships, quantitatively calculate the fracture penetration index (BCI), and abstract the fracture network into a graph model based on graph theory to construct a topology model that characterizes the spatiotemporal evolution of fractures, extract multidimensional time-series topology parameters, and obtain a multivariate topology time series dataset that reflects the evolution of coal and rock fractures.

[0077] The dataset processing and partitioning module is configured to divide the constructed multivariate topological time series dataset into training and testing sets, and to perform data normalization and outlier handling on the time series of each dimension of topological structure parameters.

[0078] The precursor recognition model building module is configured to construct a multivariate time series graph model MTAD-GAT based on preprocessed training data, forming a dual graph attention mechanism that is both feature-oriented and time-oriented, respectively modeling the structural dependencies and temporal evolution trends between topological parameters. Multi-head attention is introduced into the graph attention layer, with multiple independent attention heads calculating different attention scores. Local and global features extracted from the graph attention layer and the one-dimensional convolutional layer are fused and input into a gated recurrent unit (GRU) to deeply capture the dynamic evolution patterns and long-term dependencies of topological features. During training, the model is iteratively optimized by minimizing the loss function and combining backpropagation and gradient update mechanisms.

[0079] The instability precursor identification module is configured to input the output of the gated recurrent unit (GRU) into the prediction module for predicting the observation value at the next time step and the reconstruction module for characterizing the overall temporal distribution characteristics, respectively; to construct an anomaly score by fusing the errors of the two modules, which serves as the instability precursor index; to determine the rupture precursor interval by the temporal change of the anomaly score, and then to identify the high-contribution topological parameters and their change characteristics that have changed significantly within the precursor interval.

[0080] Compared with the prior art, the present invention has the following beneficial effects:

[0081] This invention utilizes microseismic monitoring data to invert the source mechanism, accurately capturing the source characteristics of fractures and dynamically tracking the occurrence, expansion, and penetration of fractures. This allows for the construction of fracture topology structures that more closely resemble the actual evolutionary patterns, providing a new path for achieving more physically based early warning of coal and rock mass instability and significantly improving the scientific rigor and reliability of precursor identification.

[0082] This invention constructs a fracture network topology analysis model based on graph theory. By extracting multi-dimensional topology parameters, it characterizes the evolution law and local anomaly features of the fracture network, realizing the structural identification and quantitative representation of the precursors of instability and rupture. This breaks through the limitations of traditional early warning based on signal features and provides a more accurate and interpretable intelligent early warning method for coal and rock dynamic disasters.

[0083] This invention introduces the multivariate time series graphical model MTAD-GAT, which employs an unsupervised learning mechanism. It eliminates the need for pre-defined labels or manually defined feature thresholds, automatically mining potential anomalous patterns from historical fracture topology evolution data, significantly improving the model's adaptability and generalization ability. Secondly, MTAD-GAT possesses multivariate recognition capabilities, simultaneously handling the joint evolutionary features of multiple topological parameters, avoiding misjudgments and missed detections that may arise from single indicators, thus enhancing the comprehensiveness and robustness of precursor identification. Furthermore, multi-head attention is introduced to improve the model, enabling it to capture different structural and temporal-dependent features in parallel from multiple subspaces, effectively enhancing the stability and expressive power of feature representation, further improving the perception of complex fracture evolution patterns and the accuracy of anomaly detection.

[0084] This invention fully integrates graph modeling and multivariate time series graph modeling, realizing a new path for precursor identification and disaster early warning based on fracture structure evolution. It breaks through the traditional early warning method based on signal analysis and establishes an intelligent identification framework oriented towards the fracture structure evolution mechanism. It provides more scientific and efficient technical support for early warning and proactive prevention and control of coal and rock dynamic disasters, and has significant engineering practical value and broad prospects for promotion and application.

[0085] The preferred embodiments of the invention will be described in more detail below with reference to the accompanying drawings, so as to facilitate an understanding of the features and advantages of the invention. Attached Figure Description

[0086] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. The drawings are merely illustrative of some embodiments of the present invention and are not intended to limit the scope of the present invention to all embodiments.

[0087] Figure 1 This is a flowchart of a method for identifying precursors of coal and rock instability based on fracture topology parameters according to an embodiment of the present invention;

[0088] Figure 2 This is a diagram showing four spatial topological relationships between cracks according to an embodiment of the present invention;

[0089] Figure 3 The fracture network according to an embodiment of the present invention is generalized as a graph structure;

[0090] Figure 4 This is a schematic diagram of the MTAD-GAT multivariate time series graphical model according to an embodiment of the present invention.

[0091] Figure 5 This is a feature-oriented multi-head graph attention (GAT) layer according to an embodiment of the present invention. Detailed Implementation

[0092] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The same reference numerals in the drawings represent the same components. It should be noted that the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0093] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms “first,” “second,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, “an” or “a” and similar terms do not necessarily indicate a quantity limitation. Terms such as “comprising” or “including” mean that the element or object preceding the word encompasses the element or object listed following the word and its equivalents, without excluding other elements or objects. Terms such as “connected” or “linked” are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as “upper,” “lower,” “left,” and “right” are used only to indicate relative positional relationships; these relative positional relationships may change accordingly when the absolute position of the described object changes.

[0094] To address this deficiency, monitoring and early warning methods based on fracture structure have gradually become a research hotspot. By constructing fracture topology networks and analyzing the topological changes of the network structure during temporal evolution, the correlation and synergistic anomaly characteristics between fractures can be deeply explored, identifying potential rupture precursors at the structural level, significantly improving the physical interpretability and reliability of early warning. Simultaneously, introducing neural networks to perform temporal modeling of fracture topology parameters can effectively capture the nonlinear relationships between topological features and their dynamic trends over time, improving the sensitivity and accuracy of precursor identification. Therefore, precursor identification methods based on fracture network structures and neural networks can more comprehensively and accurately reveal the evolution process of fractures within coal and rock masses, providing strong technical support for mine safety management and greatly improving the efficiency and accuracy of precursor identification.

[0095] According to a first aspect of the present invention, a method for identifying precursors of coal and rock instability based on fracture topology parameters is provided, such as... Figure 1 , Figure 4 As shown, it includes the following steps:

[0096] S10: Raw acoustic data is acquired through a microseismic monitoring system. Combined with source location calculation and source mechanism inversion, core physical parameters directly related to the micro-fracture source are obtained during the coal and rock fracturing process.

[0097] S20: Taking into account the geometric size and directional characteristics of fractures, four types of topological relationships are defined, the fracture penetration index (BCI) is quantitatively calculated, and the fracture network is abstracted into a graph model based on graph theory. A topological model that characterizes the spatiotemporal evolution of fractures is constructed, and multi-dimensional time-series topological parameters such as connection type and network density are extracted to obtain a multivariate topological time series dataset that reflects the evolution of coal and rock fractures, providing structured input data for precursor identification.

[0098] S30: Divide the constructed multivariate topological time series dataset into training and test sets, and perform data normalization and outlier removal on the time series of each dimension of topological structure parameters; data normalization is applied to both the training and test sets, while outlier removal is applied only to the training set.

[0099] S40: Based on the preprocessed training set data, a multivariate time series graph model MTAD-GAT is constructed to form a dual graph attention mechanism that is feature-oriented and time-oriented, respectively modeling the structural dependencies and temporal evolution trends between topological parameters. Multi-head attention is introduced into the graph attention layer, and different attention scores are calculated through multiple independent attention heads to obtain more stable and expressive feature representations, thereby enhancing the model's ability to model local structures and temporal patterns. The local and global features extracted from the graph attention layer and the one-dimensional convolutional layer are fused and input into the gated recurrent unit (GRU) to deeply capture the dynamic evolution patterns and long-term dependencies of topological features. During training, the model is iteratively optimized by minimizing the loss function and combining backpropagation and gradient update mechanisms.

[0100] S50: The trained model is applied to the test set data. The output of the gate-gated recurrent unit (GRU) is fed into the prediction module for predicting the observations at the next time step and the reconstruction module for characterizing the overall temporal distribution features, respectively. The errors of the two modules are fused to construct an anomaly score, which serves as an instability precursor index. The temporal changes of the anomaly score are used to determine the rupture precursor interval, thereby identifying the high-contribution topological parameters and their change characteristics that change significantly within the precursor interval.

[0101] It is understandable that, such as Figure 1 and Figure 4 As shown, the multivariate time series graphical model MTAD-GAT includes: a one-dimensional convolutional layer, a graph attention layer, a gated recurrent unit (GRU), a prediction module, and a reconstruction module. The graph attention layer includes a parallel feature-oriented graph attention GAT layer and a time-oriented graph attention GAT layer. The front end of the graph attention layer is coupled to the one-dimensional convolutional layer, and the back end of the graph attention layer is coupled to the gated recurrent unit (GRU) through concatenation. The gated recurrent unit (GRU) is coupled to the parallel prediction module and reconstruction module, respectively.

[0102] This identification method utilizes microseismic monitoring data to invert the source mechanism, accurately capturing the source characteristics of fractures and dynamically tracking the occurrence, expansion, and penetration of fractures. This allows for the construction of fracture topology structures that more closely resemble the actual evolutionary patterns, providing a new path for achieving more physically based early warning of coal and rock mass instability and significantly improving the scientific rigor and reliability of precursor identification.

[0103] This identification method is based on graph theory to construct a fracture network topology analysis model. By extracting multi-dimensional topology parameters, it characterizes the evolution law and local anomaly features of the fracture network, realizing the structural identification and quantitative representation of the precursors of instability and rupture. It breaks through the limitations of traditional early warning based on signal features and provides a more accurate and interpretable intelligent early warning means for coal and rock dynamic disasters.

[0104] This identification method introduces the multivariate time series graph model MTAD-GAT, employing an unsupervised learning mechanism. It eliminates the need for pre-defined labels or manually defined feature thresholds, automatically mining potential anomalous patterns from historical fracture topology evolution data, significantly improving the model's adaptability and generalization ability. Secondly, MTAD-GAT possesses multivariate identification capabilities, simultaneously handling the joint evolutionary features of multiple topological parameters, avoiding misjudgments and missed detections that may arise from single indicators, thus enhancing the comprehensiveness and robustness of precursor identification. Furthermore, multi-head attention is introduced to improve the model, enabling it to capture different structural and temporal-dependent features in parallel from multiple subspaces, effectively enhancing the stability and expressive power of feature representation, further improving the perception of complex fracture evolution patterns and the accuracy of anomaly detection.

[0105] This identification method fully integrates graph modeling and multivariate time series graph modeling, realizing a new path for precursor identification and disaster early warning based on the evolution of fracture structure. It breaks through the traditional early warning method based on signal analysis and establishes an intelligent identification framework oriented towards the evolution mechanism of fracture structure. It provides more scientific and efficient technical support for early warning and proactive prevention and control of coal and rock dynamic disasters, and has significant engineering practical value and broad prospects for promotion and application.

[0106] In one example of the present invention, in step S10, as Figure 2 As shown, raw acoustic data was acquired, and combined with source location calculations and focal mechanism inversion, core parameters of the coal and rock fracture source directly related to the fracture were obtained. The specific steps include the following:

[0107] S11: The raw acoustic data collected by the microseismic monitoring system is post-processed, and the arrival time and amplitude information of the first wave are extracted using the red-delay information criterion method.

[0108] S12: Using the first wave arrival time data from sensors at different locations, the spatial location calculation of the acoustic emission source is performed using the simplex positioning robust algorithm;

[0109] S13: Based on the constraints of the tension-shear rupture source model, and combined with the first wave amplitude data from sensors at different locations, solve for the six components of the source rupture moment tensor.

[0110] S14: The core physical parameters of the coal and rock fracture source are quantitatively inverted by calculating the moment tensor components. The core physical parameters include the location coordinates, spatial orientation, and fracture volume.

[0111] In one example of the present invention, in step S20, four types of topological relationships are defined by comprehensively considering the geometric size and directional characteristics of the fracture, the fracture continuity index (BCI) is quantitatively calculated, the fracture network is abstracted into a graph model based on graph theory, a topological model characterizing the spatiotemporal evolution of the fracture is constructed, multidimensional time-series topological parameters are extracted, and a multivariate topological time series dataset reflecting the evolution of coal and rock fractures is obtained, including the following steps:

[0112] S21: Based on geometric relationships, using the fracture location coordinates, volume radius, and spatial orientation, construct the normal vector and parametric equations of the intersection line between any two fracture planes, and then solve them simultaneously with the equations of the sphere containing the fracture. By analyzing the spatial positional relationship of the intersection points, determine four types of topological relationships between fractures: intersecting, connected, embedded, or disjoint. Simultaneously, calculate the diameter of the intersection circle based on the distance between the fracture spherical radius and the center of the sphere to characterize the geometric connection features of the fractures.

[0113] S22: For any fractures i and j in a fracture network, after determining the existence of a connection relationship based on the spatial topological relationship between the fractures, the formula for calculating the fracture penetration index is as follows:

[0114]

[0115] In the formula, Z(V) i ) and Z(V j ) are the normalized volume values ​​of the i-th and j-th fractures, respectively, and d ij Let be the Euclidean distance between the i-th fracture and the j-th fracture;

[0116] Assuming there are n continuous fracture nodes around fracture i, the total fracture continuity index (BCI) is... i The calculation formula is as follows:

[0117]

[0118] S23: After determining the connectivity between fractures, construct a fracture network topology model based on the geometric relationship between the fracture center and the intersection center;

[0119] S24: Extract multidimensional topological attribute parameters of the fracture network, and combine them with the time series information of microseismic events to construct a multivariate topological time series feature dataset characterizing the time-varying evolution of the fracture network structure; among which, the multidimensional topological attribute parameters include: connection type, network density, clustering coefficient, cumulative openness and connectivity index.

[0120] In one example of the present invention, in step S23, as Figure 3 As shown, the process of constructing the fracture network topology model includes the following steps:

[0121] The fracture network is abstracted into a graph structure using graph theory, where fractures are nodes and are classified into I nodes, T nodes, Y nodes, X nodes and D nodes according to the number of connecting edges. The degree (deg) of a node is defined as the total number of connecting edges.

[0122] The geometric information of the fracture is assigned to the edges of the connecting nodes. The edges are divided into unidirectional edges and bidirectional edges according to the penetration index (BCI). Among them, unidirectional edges represent lower penetration and weak interaction, while bidirectional edges represent higher penetration and strong interaction, which can form stable fracture channels.

[0123] In one example of the present invention, such as Figure 5 As shown, in step S24, multi-dimensional topological attribute parameters such as connection type, network density, clustering coefficient, cumulative openness, and connectivity index of the fracture network are extracted, specifically as follows:

[0124] ① Connection type: includes three types: node type, edge type, and edge directionality. These are used to analyze the trend of the number and proportion of 5 node types, 15 edge types (II, IT, IY, IX, ID, TT, TY, TX, TD, YY, YX, YD, XX, XD, DD), unidirectional edges, and bidirectional edges in the fracture network model as a function of coal and rock deformation.

[0125] ② Network density: The ratio of the number of edges in a fracture network to the maximum number of edges it can accommodate. It describes the edge density between nodes in the network and represents the density of the fracture distribution.

[0126]

[0127] In the formula, N e Let N be the total number of edges in the fracture network, and N be the total number of nodes.

[0128] ③ Clustering coefficients: The local clustering coefficient C(v) is the ratio of the actual number of edges between a node's neighboring nodes to the total number of possible edges between that node's neighboring nodes, reflecting the density of the local network structure around that node; the global clustering coefficient C is the average of the local clustering coefficients of all nodes, measuring the degree of clustering of all nodes in the graph, reflecting the density of the global structure of the fracture network.

[0129]

[0130] In the formula, E v k is the actual number of edges between the neighboring nodes of node v. v Let v be the degree of node v, and N be the total number of nodes in the graph.

[0131] ④ Cumulative aperture: The sum of the apertures of all nodes in the fracture network and the trend of changes in the stage of coal and rock deformation.

[0132] ⑤ Connectivity Index: The spatial distribution trend of the connectivity index of all nodes in the fracture network as the coal and rock deformation changes in stages.

[0133] In one example of the present invention, step S30, which involves partitioning the multivariate topological time series dataset and performing data preprocessing, includes the following steps:

[0134] S31: Divide the constructed multivariate topological time series dataset into a training set and a test set. The training set selects data from the stable evolution stage from the early to the middle stage of coal and rock fracturing, while the test set selects data from the later stage before instability and fracturing.

[0135] S32: To eliminate the inconsistency in the dimensions of different topological parameters, the time series of each dimension of the topological structure parameter in the training and test sets are normalized. For topological parameters x1, x2, ..., x... n-1 x n The minimum-maximum normalization method is used, and its expression is:

[0136]

[0137] In the formula, This represents the minimum value of the topology parameter. The maximum value of the topology parameter, y i ∈[0,1];

[0138] S33: The spectral residual (SR) method is used to identify and remove outliers in the time series of topological parameters of each dimension in the training set. This method extracts the frequency domain features of the time series through Fourier transform, calculates the spectral residual of its amplitude spectrum, reconstructs the anomalous enhancement signal through inverse transform, and finally sets a threshold based on the amplitude distribution of the reconstructed signal to identify and remove potential outliers.

[0139] In one example of the present invention, in step S40, the MTAD-GAT model includes a 1-D convolutional layer (kernel size 7), which is used to extract high-level features for each timestamp from the input multivariate time series data, specifically including:

[0140] Let the input tensor of the multivariate time series be X∈R. T×N×D Where T is the maximum timestamp length, N is the number of variables, and D is the input feature dimension of each variable. The feature extraction formula through a 1-D convolutional layer is as follows:

[0141] X conv =Conv1D(X)∈RT×N×D′

[0142] In the formula, X conv is the output tensor of the 1-D convolutional layer, and D′ is the dimension of the local features extracted after passing through the 1-D convolutional layer.

[0143] In one example of the present invention, in step S40, the dual-graph attention mechanism includes the following steps:

[0144] S41: Linear Transformation: For each attention head m = 1, 2, ..., M, transform the output of each node after passing through a 1-D convolutional layer.

[0145] Feature X conv ∈R T×N×D′ Perform a linear mapping, the expression of which is:

[0146]

[0147] Where: h i ∈R D′ W represents the output feature of the i-th node after passing through the convolutional layer. (m) ∈R F×D′ Let F be a learnable linear mapping matrix, and let F be the output dimension of each attention head. Features after mapping;

[0148] S42: Calculating the attention coefficient: For two nodes i and j, the attention mechanism is used to calculate their similarity. The expression for the attention coefficient is:

[0149]

[0150] In the formula: Let α be the non-normalized attention score between nodes i and j in the m-th attention head, and LeakyReLU denote the non-linear activation function. (m) ∈R 2F A learnable multi-head attention vector. This represents a vector concatenation operation;

[0151] S43: Normalized Attention Weights: The attention scores are normalized using Softmax, and the expression is as follows:

[0152]

[0153] In the formula, Let represent the normalized attention score between nodes i and j in the m-th attention head, Softmax represent the activation function, and N(i) be the set of neighboring nodes of node i.

[0154] S44: Aggregate Neighbor Features: By weighting the features of neighbor nodes through attention, the expression for updating a node is:

[0155]

[0156] In the formula, σ represents the aggregated feature output of node i in the m-th attention head, where σ is a non-linear activation function.

[0157] S45: Multi-head output aggregation: Concatenates the output features of each head along its dimensions. The expression is as follows:

[0158]

[0159] In the formula, z i This represents the final representation of node i after all heads have been merged;

[0160] The final output of the feature-oriented graph attention layer is: The final output of the time-oriented graph attention layer is The outputs of the feature-oriented graph attention layer and the time-oriented graph attention layer are arranged into a unified tensor, which is then concatenated with the output of the 1-D convolutional layer to obtain the overall representation after output concatenation. This representation is then used as input to the GRU layer for dynamic temporal modeling.

[0161]

[0162] In the formula, M·F is the dimension of the final output of each GAT layer, and D″ is the total feature dimension after concatenation.

[0163] In one example of the present invention, in step S40, a multivariate time series graphical model MTAD-GAT is introduced. The gated recurrent unit in this model is a recurrent neural network that can effectively model time series data. Its structure is suitable for capturing long-term dependencies in time series data. Specifically:

[0164] After feature extraction through 1-D convolution and GAT layers, the resulting high-dimensional features are input into the GRU layer. The output dimension of this layer is set to d1, indicating that the GRU layer will further compress and learn the temporal features of the input data, thereby capturing complex patterns and dynamic evolution laws in multiple time series.

[0165] In one example of the present invention, step S50 includes the following steps:

[0166] S51: Joint Optimization Strategy to Construct Multi-Objective Loss Function: Combining the prediction and reconstruction modules, a joint loss function is introduced during training to simultaneously optimize the two sub-networks, as shown in the following formula:

[0167] Loss = Lossfor +Loss rec

[0168] In the formula, Loss for Loss is the loss function for the prediction model. rec The loss function for reconstructing the model;

[0169] S52: Constructing an anomaly scoring mechanism: The model generates prediction error and reconstruction probability separately, and constructs a comprehensive anomaly scoring function by fusing the two. The Peak Exceeds Threshold (POT) algorithm is used to select the anomaly threshold on the validation set. If the anomaly score at a certain time step is greater than the anomaly threshold, that time step is marked as "abnormal"; otherwise, it is marked as "normal." The "abnormal" score is used as an instability precursor index to identify the rupture precursor interval.

[0170] The formula for calculating the comprehensive anomaly scoring function is as follows:

[0171]

[0172] In the formula, For predicted values Compared with the actual value x i The squared error between (1-p) i ) represents the probability of encountering an outlier in feature i based on the reconstructed model, and γ is a hyperparameter selected for the validation set to balance the two modules.

[0173] S53: Statistically analyze the topological evolution trend within the precursor interval: Statistically measure the average, variance, slope, and other characteristic quantities of each topological parameter within the precursor interval and compare them with the corresponding parameters in the normal interval; Use parameter difference analysis and t-test methods to determine whether there are significant changes in the topological parameters during the precursor stage, in order to identify high-contribution topological parameters and their change characteristics that play a key role in the rupture evolution process; If multiple parameters show synchronous or coordinated change trends within the same precursor interval, such as synchronous increase or synchronous fluctuation enhancement, it may indicate that the overall fracture network structure has entered an unstable evolution state, providing important topological precursor information for the impending rupture.

[0174] In one example of the present invention, the prediction module is configured to connect three fully connected layers with a hidden dimension of d2 after the temporal feature representation output by the gated recurrent unit (GRU) for nonlinear prediction of the data at the next time step; wherein the prediction loss function is defined using the root mean square error (RMSE) as follows:

[0175]

[0176] In the formula: x n Given the input sequence x = (x0, x1, ..., x... n-1The true value of x at the next time step. n,i Let i be the true value of the i-th feature. The predicted value is K, where K is the total number of features.

[0177] The reconstruction module is configured to learn the probability distribution of the entire input sequence in the latent space using a variational autoencoder (VAE), reconstruct the input, and calculate the reconstruction error; wherein, the reconstruction loss function is expressed as:

[0178]

[0179] In the formula, the first term is the reconstruction error (negative log-likelihood), which measures how close the reconstructed value is to the original input, q. φ (z|x) is the encoder. The first term is the decoder; the second term is the KL divergence, the regularized latent space distribution, and the latent variables. Represents the implicit structure of data.

[0180] The VAE model includes encoder q. φ (z|x) maps the input sequence to a latent variable distribution; decoder Reconstruct the input sequence based on the latent variables.

[0181] According to a second aspect of the present invention, a coal and rock instability precursor identification system based on fracture topology parameters includes:

[0182] The source parameter acquisition module is configured to acquire raw acoustic data through a microseismic monitoring system, and combine source location calculation and source mechanism inversion to obtain core physical parameters directly related to the micro-fracture source during the coal and rock fracturing process.

[0183] The topology parameter acquisition module is configured to comprehensively consider the geometric size and directional characteristics of fractures to define four types of topological relationships, quantitatively calculate the fracture penetration index (BCI), and abstract the fracture network into a graph model based on graph theory to construct a topology model that characterizes the spatiotemporal evolution of fractures. It extracts multi-dimensional time-series topology parameters such as connection type and network density to obtain a multivariate topology time-series dataset that reflects the evolution of coal and rock fractures, providing structured input data for precursor identification.

[0184] The dataset processing and partitioning module is configured to divide the constructed multivariate topological time series dataset into training and test sets, and to perform data normalization and outlier removal on the time series of each dimension of topological structure parameters. Data normalization is applied to both the training and test sets, while outlier removal is applied only to the training set.

[0185] The precursor recognition model building module is configured to construct a multivariate time series graph model MTAD-GAT based on preprocessed training data, forming a dual graph attention mechanism that is feature-oriented and time-oriented, respectively modeling the structural dependencies and temporal evolution trends between topological parameters. Multi-head attention is introduced into the graph attention layer, with multiple independent attention heads calculating different attention scores to obtain more stable and expressive feature representations, thereby enhancing the model's ability to model local structures and temporal patterns. The local and global features extracted from the graph attention layer and the one-dimensional convolutional layer are then input into a gated recurrent unit (GRU) to deeply capture the dynamic evolution patterns and long-term dependencies of topological features. During training, the model is iteratively optimized by minimizing the loss function and combining backpropagation and gradient update mechanisms.

[0186] The instability precursor identification module is configured to apply the trained model to the test set data, and use the output of the gated recurrent unit (GRU) to feed into the prediction module for predicting the observation value at the next time step and the reconstruction module for characterizing the overall temporal distribution characteristics, respectively; the error of the two modules is fused to construct an anomaly score as an instability precursor index; the temporal change of the anomaly score is used to determine the rupture precursor interval, and then the high-contribution topological parameters and their change characteristics that change significantly within the precursor interval are identified.

[0187] This identification system utilizes microseismic monitoring data to invert the source mechanism, accurately capturing the source characteristics of fractures and dynamically tracking the occurrence, expansion, and penetration of fractures. This allows for the construction of fracture topology structures that more closely resemble the actual evolutionary patterns, providing a new path for achieving more physically based early warning of coal and rock mass instability and significantly improving the scientific rigor and reliability of precursor identification.

[0188] This identification system constructs a fracture network topology analysis model based on graph theory. By extracting multi-dimensional topology parameters, it characterizes the evolution law and local anomaly features of the fracture network, realizing the structural identification and quantitative representation of the precursors of instability and rupture. It breaks through the limitations of traditional early warning based on signal features and provides a more accurate and interpretable intelligent early warning method for coal and rock dynamic disasters.

[0189] This identification system introduces the multivariate time-series graphical model MTAD-GAT, employing an unsupervised learning mechanism. It eliminates the need for pre-defined labels or manually defined feature thresholds, automatically mining potential anomaly patterns from historical fracture topology evolution data, significantly improving the model's adaptability and generalization ability. Secondly, MTAD-GAT possesses multivariate recognition capabilities, simultaneously handling the joint evolutionary features of multiple topological parameters, avoiding misjudgments and missed detections that may arise from single indicators, thus enhancing the comprehensiveness and robustness of precursor identification. Furthermore, multi-head attention is introduced to improve the model, enabling it to capture different structural and temporal-dependent features in parallel from multiple subspaces, effectively enhancing the stability and expressive power of feature representation, further improving the perception of complex fracture evolution patterns and the accuracy of anomaly detection.

[0190] This identification system fully integrates graph modeling and multivariate time series graph modeling, realizing a new path for precursor identification and disaster early warning based on the evolution of fracture structure. It breaks through the traditional early warning method based on signal analysis and establishes an intelligent identification framework oriented towards the evolution mechanism of fracture structure. It provides more scientific and efficient technical support for early warning and proactive prevention and control of coal and rock dynamic disasters, and has significant engineering practical value and broad prospects for promotion and application.

[0191] The foregoing description, with reference to preferred embodiments, details the exemplary implementation of the coal and rock instability precursor identification method and system based on fracture topology parameters proposed in this invention. However, those skilled in the art will understand that various modifications and alterations can be made to the above specific embodiments without departing from the concept of this invention, and various combinations can be made to the various technical features and structures proposed in this invention without exceeding the protection scope of this invention, which is determined by the appended claims.

Claims

1. A method for identifying a coal rock instability precursor based on a fracture topological parameter, characterized in that , comprising the following steps: S10: obtaining original acoustic wave data by a microseismic monitoring system, and obtaining core physical parameters directly related to micro-fracture sources in the coal rock fracture process by combining source location calculation and source mechanism inversion; S20: defining four types of topological relations by comprehensively considering the geometric size and direction characteristics of the fractures, quantitatively calculating the fracture breakthrough index BCI, abstracting the fracture network into a graph model based on the graph theory method, constructing a topological model reflecting the space-time evolution characteristics of the fractures, extracting multi-dimensional time-series topological parameters, and obtaining a multi-variable topological time-series data set reflecting the evolution of the coal rock fracture; S30: dividing the multi-variable topological time-series data set constructed into a training set and a test set, and performing data normalization and outlier processing on each dimension of the topological structure parameter time series; S40: based on the pre-processed training set data, constructing a multi-variable time-series graph model MTAD-GAT to form a feature-oriented and time-oriented double-graph attention mechanism, and modeling the structural dependence and time evolution trend between the topological parameters; introducing multi-head attention in the graph attention layer, calculating different attention scores through multiple independent attention heads; inputting the local and global features extracted by the graph attention layer and the one-dimensional convolution layer into the gated recurrent unit GRU to deeply capture the dynamic evolution mode and long-term dependence relationship of the topological features; in the training process, the model is iteratively optimized by minimizing the loss function, combined with the back propagation and gradient update mechanism; S50: applying the trained model to the test set data, inputting the output of the gated recurrent unit GRU into a prediction module for predicting the next time observation value and a reconstruction module for describing the overall time-series distribution characteristics; constructing an anomaly score as a precursor index by fusing the errors of the two; determining the precursor interval before the fracture by the time-series change of the anomaly score, and identifying the high-contribution topological parameters and their change characteristics in the precursor interval.

2. The coal rock instability precursor identification method based on fissure topological parameters according to claim 1, characterized in that , In the step S10, the original acoustic wave data is collected, and the core parameters of the coal rock fracture source directly related to the fracture are obtained by combining source location calculation and source mechanism inversion, specifically comprising the following steps: S11: the original acoustic wave data collected by the microseismic monitoring system is post-processed, and the first arrival time and amplitude information is extracted by using the Akaike information criterion method; S12: the spatial location calculation of the acoustic emission source is performed by using the simplex positioning robust algorithm through the first arrival time data of different position sensors; S13: based on the constraint conditions of the shear fracture source model, the six components of the source fracture moment tensor are solved by combining the first arrival amplitude data of different position sensors; S14: the core physical parameters of the coal rock fracture source are quantitatively inverted by the calculated moment tensor components, wherein the core physical parameters include the positioning coordinates, the spatial orientation and the fracture volume.

3. The coal rock instability precursor identification method based on fissure topological parameters according to claim 1, characterized in that , In the step S20, four types of topological relations are defined by comprehensively considering the geometric size and direction characteristics of the fissures, the fracture breakthrough index BCI is quantitatively calculated, the fissure network is abstracted into a graph model based on the graph theory method, the topological model reflecting the spatio-temporal evolution characteristics of the fissures is constructed, the multi-dimensional time-series topological parameters are extracted, and a multi-variable topological time-series data set reflecting the coal rock fracture evolution is obtained, including the following steps: S21: Based on the geometric relation, the normal vector and parameter equation of the intersection line of any two fissure planes are constructed by using the fissure positioning coordinates, the volume radius and the spatial orientation, and are combined with the spherical equation of the fissure; by analyzing the spatial position relation of the intersection point, the intersection, connection, embedding or separation of the four types of topological relations between the fissures is determined; meanwhile, according to the distance between the fissure spherical radius and the spherical center, the diameter of the intersection line circle is calculated to represent the geometric connection characteristics of the fissures; S22: For any fracture in the fracture network i and j , after determining the connection relationship according to the spatial topological relationship between the fractures, the fracture penetration index is calculated as follows: wherein and are the normalized volume values of the first i and second j fractures, respectively, is the Euclidean distance between the first i and second j fractures. Assuming fissure i There are n The total fracture penetration index The formula is as follows: S23: After determining the connection relation between the fissures, the fissure network topological model is constructed based on the geometric relation between the fissure center and the intersection line center; S24: The multi-dimensional topological attribute parameters of the fissure network are extracted, the time-series information of the microseismic events is combined, a multi-variable topological time-series characteristic data set reflecting the time-varying evolution characteristics of the fissure network structure is constructed; wherein the multi-dimensional topological attribute parameters include: connection type, network density, clustering coefficient, cumulative opening degree and breakthrough index.

4. The coal rock instability precursor identification method based on fissure topological parameters according to claim 1, characterized in that , In the step S23, the fissure network topological model construction process includes the following steps: The fissure network is abstracted into a graph structure by using the graph theory method, wherein the fissures are taken as nodes, the nodes are divided into I nodes, T nodes, Y nodes, X nodes and D nodes according to the number of connection edges, and the degree of the node is defined as the total number of connection edges; The geometric information of the fissures is distributed to the edges of the connected nodes, and the edges are divided into one-way edges and two-way edges according to the breakthrough index BCI; wherein the one-way edge represents lower breakthrough and weak interaction, and the two-way edge represents higher breakthrough and strong interaction, and can form a stable fissure channel.

5. The coal rock instability precursor identification method based on fissure topological parameters according to claim 1, characterized in that , In the step S30, the multi-variable topological time-series data set is divided and data preprocessed, including the following steps: S31: The multi-variable topological time-series data set constructed is divided into a training set and a test set, wherein the training set selects the stable evolution stage data from the early to middle stage of the coal rock fracture evolution, and the test set selects the late stage data near the pre-stability of the fracture; S32: Normalization is performed on each dimension of the time series of the topology structure parameters in the training set and the test set. For the topology parameters , the minimum-maximum normalization method is adopted, and the expression is: wherein is the minimum value of the topology parameter, is the maximum value of the topology parameter, yi ∈ [0, 1]; S33: The spectral residual method is used to identify and remove the abnormal points in the time series of each dimensional topological structure parameter in the training set.

6. The coal rock instability precursor identification method based on fissure topological parameters according to claim 1, characterized in that , In the step S40, the MTAD-GAT model includes a 1-D convolution layer, which is used to extract high-level features of each time stamp from the input multi-variable time series data, including: Let the multivariate time series input tensor be , where T is the maximum timestamp length, N is the number of variables, D is the input feature dimension of each variable, and the feature extraction is performed by a 1-D convolutional layer with the following formula: In the formula, is a 1-D convolutional layer output tensor, is a local feature dimension extracted after the 1-D convolutional layer.

7. The coal rock instability precursor identification method based on fissure topological parameters according to claim 1, characterized in that , In the step S40, the double-graph attention mechanism includes the following steps: S41: linear transformation: for each attention head , the output features of each node passing through the 1-D convolution layer are linearly mapped, and the expression is: In the formula: For the first i The output features of each node after passing through the convolutional layer, For a learnable linear mapping matrix, F For each attention head's output dimension, Features after mapping; S42: Calculate the attention coefficient: for two nodes i and j, the attention mechanism is used to calculate the similarity, and the expression of the attention coefficient is: In the formula: For the first m Nodes in each attention head i and j The nonnormalized attention score between them LeakyReLU Represents a nonlinear activation function. For learnable multi-head attention vectors, This represents a vector concatenation operation; S43: Normalizing attention weights: Use Softmax The attention scores are normalized, with the expression: In the formula, For the first m Nodes in each attention head i and j Normalized attention scores between Softmax Represents the activation function. For nodes i The set of adjacent nodes; S44: Aggregate neighbor features: update the expression of the node by weighting the features of the neighbor nodes through attention, and the expression is: wherein, is the aggregated feature output of the i-th attention head node m i is a non-linear activation function.​​ S45: Multi-head output aggregation: concatenate the output features of each head along the dimension, which can be expressed as: wherein is the final representation of the fused node for all heads i . wherein the feature-oriented graph attention layer finally outputs as the time-oriented graph attention layer finally outputs as ; the output of the feature-oriented graph attention layer and the output of the time-oriented graph attention layer are arranged into a unified tensor, and are spliced with the output of the 1-D convolution layer to obtain an output spliced total representation, so as to input a GRU layer to perform dynamic time sequence modeling: In the formula, is the dimension of the final output of each GAT layer, is the total feature dimension after splicing.

8. The coal rock instability precursor identification method based on fissure topological parameters according to claim 1, characterized in that , The step S50 includes the following steps: S51: Joint optimization strategy to build multi-objective loss function: combine the prediction module and the reconstruction module, and introduce a joint loss function in the training process to optimize the two sub-networks simultaneously, which can be expressed as: In the formula, is a loss function of the prediction model, is a loss function of the reconstruction model; S52: Build anomaly scoring mechanism: the model generates prediction error and reconstruction probability, respectively, and builds a comprehensive anomaly scoring function by fusing the two. The peak value exceeding threshold algorithm is used to select the anomaly threshold on the validation set. If the anomaly score of a certain time step is greater than the anomaly threshold, the time step is marked as "abnormal", otherwise it is marked as "normal". The "abnormal" score is used as the precursor index to identify the precursor interval before the rupture; The calculation formula of the comprehensive anomaly scoring function is as follows: where, is the predicted value is the actual value is the squared error between the two, is the probability of encountering an anomaly for a feature i according to the reconstruction model, is the hyperparameter chosen for the validation set to balance the two modules introduced; S53: Statistics and analysis of the topological evolution trend in the precursor interval: statistics the characteristic quantities of each topological parameter in the precursor interval, and compare with the corresponding parameters in the normal interval; through parameter difference analysis and t-test method, it is judged whether there is significant change of topological parameters in the precursor stage, so as to identify the high contribution topological parameters and their change characteristics which play a key role in the evolution process of rupture.

9. The coal rock instability precursor identification method based on fissure topological parameters according to claim 1, characterized in that , The prediction module is configured to access a three-layer fully connected layer with a hidden dimension of d2 after outputting the time sequence feature representation of the gated recurrent unit (GRU), for nonlinear prediction of data of a next time step; wherein the prediction loss function is defined as The root mean square error is defined as follows: wherein: is the true value of the input sequence at the next time step, is the true value of the i th feature, is the predicted value, K is the total number of features; The reconstruction module is configured to learn the probability distribution of the entire input sequence in the latent space by using a variational autoencoder, reconstruct the input, and calculate the reconstruction error; wherein the reconstruction loss function is represented as: where the first term is the reconstruction error, measuring how close the reconstructed value is to the original input, is the encoder, is the decoder; the second term is the KL divergence, regularizing the latent space distribution, the latent variable represents the hidden structure of the data.

10. A coal rock instability precursor identification system based on fracture topological parameters, characterized in that , including: The source parameter acquisition module is configured to obtain the original acoustic data by microseismic monitoring system, and to obtain the core physical parameters directly related to micro fracture source in the process of coal rock rupture by combining source location calculation and source mechanism inversion; The topological parameter acquisition module is configured to define four types of topological relationships by comprehensively considering the geometric size and direction characteristics of the crack, to quantitatively calculate the crack breakthrough index BCI, and to abstract the crack network into a graph model based on graph theory, to build a topological model reflecting the temporal and spatial evolution characteristics of the crack, to extract multi-dimensional time series topological parameters, and to obtain the multivariate topological time series data set reflecting the evolution of coal rock rupture; The data set processing and division module is configured to divide the constructed multivariate topological time series data set into training set and test set, and to perform data normalization and outlier processing on each dimensional topological structure parameter time series; The precursor identification model construction module is configured to construct a multivariate time series graph model MTAD-GAT based on the preprocessed training set data, to form a double graph attention mechanism facing features and time, to model the structural dependence and temporal evolution trend between topological parameters respectively, to introduce multi-head attention in the graph attention layer, to calculate different attention scores through multiple independent attention heads respectively, to input the local and global features extracted by the graph attention layer and the one-dimensional convolution layer into the gated recurrent unit GRU, to deeply capture the dynamic evolution mode and long-term dependence relationship of topological features, and to optimize the model through the minimization of loss function, combined with the back propagation and gradient update mechanism during the training process. The instability precursor identification module is configured to apply the trained model to test set data, utilize the gated recurrent unit (GRU) output to be respectively transmitted into a prediction module for predicting the next time observation value and a reconstruction module for describing the overall time sequence distribution characteristics, fuse the errors of both to construct an anomaly score as an instability precursor index, and determine the precursor interval before the rupture through the time sequence change of the anomaly score, and further identify the high contribution topological parameters and their change characteristics that have significant changes in the precursor interval.

Images